June 6, 2011

The Word’s Out
By Selena Chavis
For The Record
Vol. 23 No. 11 P. 20

A recent KLAS report breaks down how speech recognition technology has become a solid option for healthcare organizations looking to save money and gain efficiencies.

Which technology will produce the greatest return on investment (ROI) in terms of streamlining processes, cutting costs, and improving patient outcomes?

This is the million-dollar question for the majority of hospitals in today’s lean market. For numerous facilities across the country, the answer has been speech recognition technology.

Consider UNC Health Care, a network of hospitals and clinics in North Carolina encompassing a 708-bed facility and the nationally ranked University of North Carolina School of Medicine. Faced with rising transcription expenses, the HIM department received a mandate to reduce the cost of its transcription services while improving turnaround time and upgrading its technology.

The healthcare network went from manual transcription to deploying a speech recognition workflow model from a leading vendor and realized an ROI of approximately $3.5 million during the first three years, according to hospital officials.

It’s the kind of success any healthcare system would hope to achieve in this era of tighter budgets and meaningful use requirements. And while not every facility that deploys speech recognition will realize millions of dollars in savings, the benefits associated with the technology are turning heads throughout the industry.

“Speech Recognition 2010,” a report from the research group KLAS, revealed that one in four hospitals now use speech recognition and that the market is ripe for rapid growth. According to the report, “Speech recognition’s influence continues to grow as speech systems become more and more common in hospitals, diagnostic imaging groups, and other outpatient care sites. Once a novel technology, speech recognition today has earned a solid place in the elite club of healthcare-IT tools with demonstrable ROI. That position is more critical than ever as the healthcare industry focuses on HITECH/meaningful use (MU), clinical IT applications like physician documentation, and improved efficiency.”

Of the survey’s 355 respondents, more than 90% were hospitals.

Ben Brown, general manager of imaging informatics with KLAS and author of “Speech Recognition 2010,” points out that while industry factors such as meaningful use are driving the growth of speech recognition, the technology has also reached a level of market saturation that it is primed for increased growth potential based on typical product life cycles.

“When you see technology adoption curves … once [a product] reaches 10% to 25% adoption in the market, then the uptake goes much faster,” he explains. “The speed of adoption increases as the percentage of industry adoption increases.”

Carina Edwards, senior vice president of healthcare marketing at Nuance Healthcare, says speech recognition will be instrumental in helping hospitals achieve the overriding goal of meaningful use: to document the patient care process in a way that makes information sharing available as quickly as possible rather than two days later, which is often the case when traditional transcription methods are employed.

“As all facilities drive to meet meaningful use, we are seeing a huge adoption of speech recognition on the front end and back end,” she says.

Moving From Back to Front
Many speech recognition deployments have started with a back-end solution, according to Edwards, adding that such a model is typically the easiest way to introduce the technology into physician workflow, especially if users are reluctant to try a new process.

With a back-end solution, the speech recognition function tends to be transparent. In other words, physicians may not even be aware that the technology is being used because they continue to dictate their notes as they had in the past. The speech recognition server automatically processes final dictations, and the transcriptionist is presented with transcribed text and the original audio file.

The transcriptionist’s main role becomes that of checking the accuracy of speech recognition rather than transcribing an entire report. “Transcriptionists now become editors,” says Brown, who references long-time industry predictions that transcriptionists would evolve into this role as speech recognition became more mainstream.

The benefit of back-end technology is that it offers efficiencies without affecting physician workflow. Sharon Barnicle, HIM director at Springhill Memorial Hospital in Mobile, Ala., says her facility’s installation of back-end speech recognition technology in 2005 had little effect on physician workflow other than a change in the phone number they dialed to dictate.

“We were able to reduce our costs of transcribed documents through the transition to edited documents,” she says. “We saw the shift and realized an initial return within six months to a year.”

While back-end programs produced the desired cost-efficiencies and ROI that Springhill was seeking, Barnicle points out that front-end speech applications will best position hospitals to meet meaningful use requirements. The front-end approach requires physicians to speak and self-edit their text directly into an EHR during the patient encounter. Different from traditional dictation, these systems can cause unwanted disturbances to physician workflow, at least in the early stages of deployment.

While front-end speech recognition has the potential to dramatically impact physician workflow, it also has the capability to cut turnaround times to a matter of minutes, according to Brown. Front-end speech systems reduce the added time associated with transcriptionists having to track down a physician or a report as well as the number of transcriptionists needed to complete editing tasks.

“It is not uncommon for a facility to reduce costs by $4,000 to $5,000 per month after adopting speech recognition,” Brown notes, pointing to some ROI as high as $45,000 per month depending on the size and scope of the healthcare facility or network. “Almost all the vendors we have talked to have proof statements like that.”

In an age when the majority of healthcare facilities are becoming more electronic, efficiency is foremost on healthcare executives’ minds. A Nuance white paper points out that front-end speech recognition offers a proven method for reducing the amount of time required for a physician to “type” documentation into an EHR. Studies reveal that the average physician spends up to 15 hours each week documenting, and the average case takes three to four times longer to document directly into the EHR by keyboard when compared to dictation. Speech recognition minimizes the time that would be associated with keyboard strokes.

With meaningful use driving the need to create efficiencies associated with turnaround times for clinical notes and documentation sharing, Springhill Memorial began deploying front-end systems last year, according to Barnicle. “The expectation now is to move it more into the electronic environment rather than traditional dictation,” she explains.

The first stop for front-end deployment was the emergency department (ED). Barnicle says the transition went much easier than expected. “They like it quite a bit. They find it more difficult to edit documents later after the patient is gone,” she says. “You would have thought with the [ED], it wouldn’t have been that way since they are so busy.”

Because of the large amount of dictation it produces, the radiology department is next on the list to undergo the transition at Springhill. Barnicle says more deployments will take place as the hospital moves from a hybrid environment to a fully electronic setting.

Edwards suggests that this sort of hybrid situation, in which front-end and back-end systems coexist, will be the norm for the next few years as facilities move to more electronic record-keeping systems. “We see a mix of adoption [of front-end technology]. Some are early adopters and want to dive in. Others are laggards and want to see how it affects workflow first,” she says. “For the next four to five years, we’ll see this need for both front-end and back-end speech.”

What the Industry Wants
Brown points to three primary factors that will help drive further expansion of speech recognition technology: meaningful use and the need to populate EHRs in a timely manner; the fact that speech recognition technology has a known value proposition and a proven track record of excellent ROI rates; and improvements in response time to patient needs and overall care thanks to increased efficiencies.

These factors will naturally drive growth, but vendors that will capture market share going forward are those that adequately respond to industry needs, Brown says. As detailed in the KLAS report, one of the concerns named by industry adopters was the continued desire for better small word and accent recognition.

“These tools are only as good as the recognition engines running through them,” Brown notes. “You have to train the system to recognize your voice.”

Issues concerning accuracy were long a primary roadblock to widespread adoption, but technological advances have paved the way to much higher levels of performance. According to Edwards, the “speech accuracy” of out-of-the-box Nuance solutions runs between 90% and 95%, and the technology continues to be refined.

Barnicle notes that the recognition of various dialects was an early issue at Springhill, but physicians were trained over time to teach the technology how to recognize speech patterns and words.

The KLAS survey revealed that providers view speech recognition technology as disruptive in both the positive and negative sense of the word. It changes workflow to be better aligned with meaningful use goals, but it also disrupts workflow on several levels.

Industry adopters would like to see improvements that would negate workflow disturbances to transcriptionists being taught how to edit documents produced by back-end systems. Also, they expressed a desire to simplify the transition for physicians being trained on front-end systems.

Edwards mentions industry efforts to move seamlessly toward offering a tool that works on both the front end and back end. “The physician just wants to dictate and be done,” she says, adding that technological advances will focus on “quicker, efficient, more comprehensive” solutions. Working alongside EHR vendors to ensure workflow is efficient will also help improve integration.

Brown says vendors can differentiate themselves by improving the provider experience during the implementation and training phases. “Also, it is difficult to overestimate the positive impact vendors can have when they deliver ongoing support with the aim of minimizing roadblocks and excuses,” he noted in the report.

The ability to introduce front-end systems at the same time as electronic clinical documentation will minimize change and disruption, Edwards adds. “When [hospitals] are training doctors, they are putting a microphone in their hand at the same time. It becomes much more natural then,” she says.

Looking ahead, the KLAS report pointed to natural language processing (NLP) functions as a next step for speech recognition technology. One of the key concerns with speech recognition has centered on its limitations with unstructured text. As advances in this area continue, speech recognition technology will be able to effectively extract from freely dictated text directly into an EHR’s appropriate data fields. The solution will be able to understand, automate, and structure data.

Brown says industry players are investing in NLP speech-driven software for clinical documentation and other applications. Advances in this area will continue to drive the growth of speech recognition.

Improvement in patient care and customer service will figure into the adoption equation, according to the KLAS report, especially as healthcare networks shift to more specialized outpatient sites. One healthcare organization that participated in the survey noted, “We get a lot more business because our stroke center can guarantee final results in 15 minutes from the time of the exam,” a feat not possible before the adoption of speech recognition.

— Selena Chavis is a Florida-based freelance journalist whose writing appears regularly in various trade and consumer publications.